nlin0505054/text.tex
1: \documentclass[letterpaper,10pt]{article}
2: \usepackage{amsmath,mathptm,graphicx,eprint}
3: 
4: \setreportheading{Masahiko Yoshioka}{PREPRINT}
5: 
6: \begin{document}
7: 
8: \title{Chaos synchronization in gap-junction-coupled neurons}
9: {\date{{June 10, 2004}\\ {\normalsize ({Revised on May 9, 2005})}}}
10: \author{{{Masahiko Yoshioka}}\thanks{Electronic address: {{myosioka@brain.riken.go.jp}}}\\
11: \begin{minipage}{1.0\textwidth}{\normalsize\it\begin{center}\ \\ {{%
12: Brain Science Institute, The Institute of Physical and Chemical Research (RIKEN)\\
13: Hirosawa 2-1, Wako-shi, Saitama, 351-0198, Japan}}\end{center}}\end{minipage}}
14: 
15: \maketitle
16: 
17: \begin{abstract}
18: 
19: Depending on temperature the modified Hodgkin-Huxley (MHH) equations
20: exhibit a variety of dynamical behavior including intrinsic chaotic
21: firing. We analyze synchronization in a large ensemble of MHH neurons
22: that are interconnected with gap junctions. By evaluating tangential
23: Lyapunov exponents we clarify whether synchronous state of neurons is
24: chaotic or periodic. Then, we evaluate transversal Lyapunov exponents
25: to elucidate if this synchronous state is stable against infinitesimal
26: perturbations. Our analysis elucidates that with weak gap junctions,
27: stability of synchronization of MHH neurons shows rather complicated
28: change with temperature.  We, however, find that with strong gap
29: junctions, synchronous state is stable over the wide range of
30: temperature irrespective of whether synchronous state is chaotic or
31: periodic. It turns out that strong gap junctions realize the robust
32: synchronization mechanism, which well explains synchronization in
33: interneurons in the real nervous system.
34: 
35: \end{abstract}
36: 
37: {
38: 
39: \vspace{0.2em}
40: 
41: \noindent {\footnotesize PACS numbers: {87.18.Sn,02.30.Oz,05.45.Xt,05.45.Ra}}
42: 
43: \vspace{1em}
44: 
45: }
46: 
47: \noindent In the rat hippocampus interneurons show the high frequency synchronization
48: during the gamma oscillation ($\sim$40Hz) and sharp wave burst
49: ($\sim$200Hz)\cite{buzsaki3}, and such simple synchronization of a large
50: ensemble of neurons has attracted much attention of theoretical
51: researchers\cite{wang,golomb,whittington1,wang2,chow,tiesinga,myosioka7,ermentrout,hansel}.
52: One major analysis for these studies is the phase reduction
53: method\cite{kuramoto,ermentrout,hansel}, in which phase variables are
54: utilized to represent the periodic behavior of neurons.  The phase
55: reduction method is, however, applicable only to the case of
56: infinitesimal interactions.  Moreover, if neurons behave aperiodic, phase variables are indefinable.  The general synchronization properties of
57: strongly coupled neurons thus remained unclear, especially in the case
58: of chaotic neurons.
59: 
60: Meanwhile, studies of synchronization of a large ensemble of chaotic
61: oscillators have made a remarkable progress in recent
62: years\cite{fujisaka,kaneko,maistrenko,pikovsky}.  The major targets of
63: these studies are simple chaotic oscillators such as Lorenz equations
64: and logistic maps.  Synchronous state of these oscillators is
65: characterized by two types of Lyapunov exponents: tangential Lyapunov
66: exponents and transversal Lyapunov exponents.  While tangential Lyapunov
67: exponents clarify whether synchronous state is chaotic or periodic,
68: transversal Lyapunov exponents elucidate if synchronous state is stable
69: against infinitesimal perturbations.  In the present study, we employ
70: these sophisticated techniques in chaos synchronization theory
71: to investigate synchronization of neurons.  We
72: show that tangential and transversal Lyapunov exponents enable us to analyze stability of
73: synchronization in a large ensemble of neurons for arbitrary neuron
74: dynamics and arbitrary strength of interactions.
75: 
76: The concrete target of the present analysis is a network of $N(\ge 2)$
77: spiking neurons that obey the modified Hodgkin-Huxley (MHH)
78: equations\cite{braun,feudel}.  The MHH equations are four-dimensional
79: nonlinear differential equations, which include temperature-dependent
80: scaling factors $\rho=A_1^{(T-T_0)/10}$ and $\phi=A_2^{(T-T_0)/10}$ (See Ref.~\cite{feudel}.)
81: With $T$ changing, a MHH neuron shows a variety of
82: dynamical behavior including chaotic firing as shown in
83: Fig.~\ref{fig:result}{(a)} (ISI and so on will be explained later.)
84: For the sake of simplicity, we denote this MHH neuron dynamics by
85: $d{{\bf x}}/dt={{\bf F}}({{\bf x}})$ with neuron state vector
86: ${{\bf x}}={\left ( {v,w_{1},\ldots,w_{n-1}} \right )}^{\rm T}$, where $v$ represents the
87: membrane potential and ${\left\{ {w_l} \right \}}$ describe gating of ion channels.
88: We assume that $N$~neurons ${\left\{ {{{\bf x}}_i} \right \}}$ are interconnected with
89: all-to-all gap junctions.  Since gap junctions induce electric currents
90: proportional to potential difference between neurons, the dynamics of
91: the neural networks is expressed as
92: \begin{equation}
93: \frac{d{{{\bf x}}}_i}{dt}={{\bf F}}({{\bf x}}_i)+(I_i/c,0,\ldots,0)^{\rm T},\ i=1,\ldots,N \label{eq:mhh}
94: \end{equation}
95: with
96: \begin{equation}
97:  I_i=\frac{g}{N}\sum_j{\left ( {v_j-v_i} \right )}=\frac{g}{N}\sum_j{\left ( {x_{j1}-x_{i1}} \right )},\label{eq:interaction}
98: \end{equation}
99: where constant $c=1.0\ \mu {\rm F}/{\rm cm}^2$ is capacitance of the
100: membrane and $I_i$ is the electric current induced by gap junctions.
101: 
102: The above dynamics can be generalized to the form
103: \begin{equation}
104: \frac{d{{{\bf x}}}_i}{dt}={{\bf F}}({{\bf x}}_i)+\frac{g}{N}\sum_j{{\bf G}}({{\bf x}}_i,{{\bf x}}_j).\label{eq:general}
105: \end{equation}
106: Therefore, we investigate synchronous state in this general mean-field
107: dynamics.  We assume stationary synchronous state
108: ${{\bf x}}_1^\ast=\ldots={{\bf x}}_N^\ast={{\bf x}}^\ast$, which obeys
109: \begin{equation}
110: \frac{d {{{\bf x}}}^\ast}{dt}={{\bf F}}({{\bf x}}^\ast)+g{{\bf G}}({{\bf x}}^\ast,{{\bf x}}^\ast).\label{eq:sync}
111: \end{equation}
112: To elucidate the stability of this synchronous state we investigate
113: perturbed state ${{\bf x}}_i={{\bf x}}^\ast+\delta{{\bf x}}_i$.
114: We define Jacobi matrices such that
115: ${{\bf F}}({{\bf x}}^\ast+\delta{{\bf x}})={{\bf F}}({{\bf x}}^\ast)+{{\bf F}}^\prime({{\bf x}}^\ast)\delta{{\bf x}}+(\mbox{higher
116: order})$ and
117: ${{\bf G}}({{\bf x}}^\ast+\delta{{\bf x}}_1,{{\bf x}}^\ast+\delta{{\bf x}}_2)={{\bf G}}({{\bf x}}^\ast,{{\bf x}}^\ast)+{{\bf G}}^\prime_1({{\bf x}}^\ast,{{\bf x}}^\ast)\delta{{\bf x}}_1+{{\bf G}}^\prime_2({{\bf x}}^\ast,{{\bf x}}^\ast)\delta{{\bf x}}_2+(\mbox{higher
118: order})$.
119: Then, Taylor series expansion to the first order yields
120: \begin{eqnarray}
121:  \frac{d(\delta{{{\bf x}}}_i)}{dt}&=&{\left [ {{{\bf F}}^\prime({{\bf x}}^\ast)+g{{\bf G}}^\prime_1({{\bf x}}^\ast,{{\bf x}}^\ast)} \right ]}\delta{{\bf x}}_i\nonumber\\
122:  &&+g{{\bf G}}^\prime_2({{\bf x}}^\ast,{{\bf x}}^\ast)\frac{1}{N}\sum_j\delta{{\bf x}}_j.\label{eq:perturbation}
123: \end{eqnarray}
124: The naive evaluation of this $N$-body dynamics brings about an eigenvalue
125: problem with the large size of matrix.
126: We hence define mean state
127: $\overline{{{\bf x}}}=(1/N)\sum_i{{\bf x}}_i$ and obtain the dynamics of its deviation~$\delta\overline{{{\bf x}}}=(1/N)\sum_i\delta{{\bf x}}_i$ in the closed form
128: \begin{equation}
129:  \frac{d(\delta{\overline{{{\bf x}}}})}{dt}={\left [ {{{\bf F}}^\prime({{\bf x}}^\ast)+g{{\bf G}}^\prime_1({{\bf x}}^\ast,{{\bf x}}^\ast)+g{{\bf G}}^\prime_2({{\bf x}}^\ast,{{\bf x}}^\ast)} \right ]}\delta\overline{{{\bf x}}}.\label{eq:dave}
130: \end{equation}
131: For this $n$-dimensional linear dynamics, we can
132: define the spectrum of $n$~Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}_{l=1,\ldots,n}$. These exponents are the so-called
133: tangential Lyapunov exponents.
134: To the first order, Eqs.~(\ref{eq:sync}) and (\ref{eq:dave}) are equivalent to
135: \begin{equation}
136: \frac{d}{dt}{({{\bf x}}^\ast+\delta\overline{{{\bf x}}})}={{\bf F}}({{\bf x}}^\ast+\delta\overline{{{\bf x}}})+g{{\bf G}}({{\bf x}}^\ast+\delta\overline{{{\bf x}}},{{\bf x}}^\ast+\delta\overline{{{\bf x}}}).\label{eq:ave}
137: \end{equation}
138: Solving Eqs.~(\ref{eq:sync}) and (\ref{eq:ave}) numerically we can
139: calculate time evolution of sufficiently small
140: deviation~$\delta\overline{{{\bf x}}}$.  Evaluating this time evolution of $\delta\overline{{{\bf x}}}$ by the
141: well-known computational method for Lyapunov exponents\cite{shimada} we
142: can calculate ${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ numerically.
143: Note that in this calculation of ${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ we do not have to solve the huge $N$-body dynamics in Eq.~(\ref{eq:general}).
144: Since replacement of ${{\bf x}}^\ast$ in
145: Eq.~(\ref{eq:sync}) by ${{\bf x}}^\ast+\delta\overline{{{\bf x}}}$ gives the same
146: dynamics as Eq.~(\ref{eq:ave}), ${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ indicate the characteristics
147: of synchronous state~${{\bf x}}^\ast$, that is, when
148: synchronous state~${{\bf x}}^\ast$ is periodic (chaotic), the largest tangential Lyapunov exponent~${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ takes the zero value (a positive value).
149: 
150: We have evaluated mean state~$\overline{{{\bf x}}}$ by tangential Lyapunov
151: exponents~${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$. We now
152: investigate 
153: deviations around mean state: ${{\bf x}}_i=\overline{{{\bf x}}}+\delta\tilde{{{\bf x}}}_i$.
154: Subtracting Eq.~(\ref{eq:dave}) from Eq.~(\ref{eq:perturbation}) we
155: obtain the dynamics of $\delta\tilde{{{\bf x}}}_i$ in the closed form
156: \begin{equation}
157:  \frac{d(\delta{\tilde{{{\bf x}}}}_i)}{dt}={\left [ {{{\bf F}}^\prime({{\bf x}}^\ast)+g{{\bf G}}^\prime_1({{\bf x}}^\ast,{{\bf x}}^\ast)} \right ]}\delta\tilde{{{\bf x}}}_i.\label{eq:dtilde}
158: \end{equation}
159: When synchronization is stable, all deviations
160: ${\left\{ {\delta\tilde{{{\bf x}}}_i} \right \}}$ must converge to ${\left\{ {{\bf 0}} \right \}}$.
161: Since these $N$ dynamics of deviations are completely identical, it
162: suffices to evaluate only one dynamics among them.  For $n$-dimensional
163: linear dynamics in Eq.~(\ref{eq:dtilde}) we can define the spectrum of
164: $n$~Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}_{l=1,\ldots,n}$. These exponents are the
165: so-called transversal Lyapunov exponents.  The largest transversal
166: Lyapunov exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ takes a negative value when synchronous
167: state is stable in the sense of Milnor\cite{maistrenko,milnor}.  To the
168: first order, Eqs.~(\ref{eq:sync}) and (\ref{eq:dtilde}) are equivalent
169: to
170: \begin{equation}
171: \frac{d}{dt}{({{\bf x}}^\ast+\delta\tilde{{{\bf x}}}_i)}={{\bf F}}({{\bf x}}^\ast+\delta\tilde{{{\bf x}}}_i)+g{{\bf G}}({{\bf x}}^\ast+\delta\tilde{{{\bf x}}}_i,{{\bf x}}^\ast).\label{eq:tilde}
172: \end{equation}
173: Applying the computational method for Lyapunov exponents\cite{shimada} to 
174: Eqs.~(\ref{eq:sync}) and (\ref{eq:tilde}), we can calculate ${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}$ numerically.
175:  
176: Let us apply the above analysis to investigating synchronization in
177: networks of MHH neurons defined by Eqs.~(\ref{eq:mhh}) and (\ref{eq:interaction}). First, we 
178: calculate synchronous state ${{\bf x}}^\ast$ in Eq.~(\ref{eq:sync}).
179: Note that in the present system the interaction term
180: $g{{\bf G}}({{\bf x}}^\ast,{{\bf x}}^\ast)$ in Eq.~(\ref{eq:sync}) vanishes because of
181: Eq.~(\ref{eq:interaction}).
182: Therefore, the behavior of ${{\bf x}}^\ast$ in Eq.~(\ref{eq:sync}) is
183: completely the same as that of an isolated single MHH neuron.
184: For the
185: rough illustration of a single MHH neuron behavior, we define
186: the $k$-th spike timing $t(k)$ by the time when membrane potential
187: $v^\ast=x_1^\ast$ crosses the threshold value $\theta=-20$~mV from
188: below, and then calculate interspike intervals (ISIs) $t(k+1)-t(k)\
189: (k=1,2,\ldots)$ in Fig.~\ref{fig:result}{(a)}.  Below $T=6.8${\char'27\kern-.3em\hbox{C}}, ISIs
190: take the single value around 650~msec, implying the simple periodic
191: firing in which only one spike arises during the period.  At $T=6.8${\char'27\kern-.3em\hbox{C}},
192: however, period doubling bifurcation occurs so that the neuron fires
193: twice during the period.  After that, following typical period doubling
194: cascade, the MHH neuron dynamics reaches the chaotic regime beyond
195: $T=7.3${\char'27\kern-.3em\hbox{C}}, where ISI distribution becomes blurred.  In this chaotic
196: regime, we, however, observe several periodic windows, in which the
197: behavior of neuron becomes periodic abruptly.
198: 
199: Second, from Eqs.~(\ref{eq:sync}) and (\ref{eq:ave}), we calculate the
200: tangential Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ for the exact characterization of
201: synchronous state~${{\bf x}}^\ast$ illustrated in
202: Fig.~\ref{fig:result}{(a)}. In the present system, ${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ are
203: independent of $g$ because of the interaction in
204: Eq.~(\ref{eq:interaction}).  In Fig.~\ref{fig:result}{(b)}, we plot the
205: largest tangential Lyapunov exponents~${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ as a function of
206: temperature $T$.  When synchronous state~${{\bf x}}^\ast$ in
207: Fig.~\ref{fig:result}{(a)}\ is periodic, ${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ takes the zero
208: value.  In chaotic regime beyond $T=7.3{\char'27\kern-.3em\hbox{C}}$, ${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ takes a positive
209: value, though we observe the several valleys of ${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ corresponding
210: to the periodic windows observed in Fig.~\ref{fig:result}{(a)}.
211: 
212: Third, we calculate the transversal Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}$
213:  from Eqs.~(\ref{eq:sync}) and (\ref{eq:tilde}).  
214: ${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}$ depend on parameter~$g$.
215: In Fig.~\ref{fig:result}{(c)}\ 
216: we calculate the largest transversal Lyapunov exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ as a
217: function of temperature~$T$ for $g=0.02$~{$\mbox{mS/cm}^2$}.  When ${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ takes a
218: negative value, synchronization of MHH neurons can occur.
219: 
220: When we assume weak gap junctions as in Fig.~\ref{fig:result}($g=0.02$~{$\mbox{mS/cm}^2$}),
221: the condition for synchronization of MHH neurons is rather complicated.
222: Periodic synchronous state is stable in some conditions and unstable in
223: other conditions. We also see stable chaotic synchronous state in
224: some values of temperature~$T$.
225: Around $T\sim 12${\char'27\kern-.3em\hbox{C}}\ we find unstable periodic synchronous state inside
226: the periodic window (${\lambda^{\mbox{\tiny $\parallel$}}_{1}}=0$ and $0<{\lambda^{\mbox{\tiny $\perp$}}_{1}}$ at $T=11.9${\char'27\kern-.3em\hbox{C}}) while we find stable chaotic synchronous state
227: outside the periodic window ($0<{\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ and ${\lambda^{\mbox{\tiny $\perp$}}_{1}}<0$ at $T=12.1${\char'27\kern-.3em\hbox{C}}). Actually, 
228: the numerical simulations of 100 MHH neurons in Fig~\ref{fig:simulation}
229: show the good agreement with the results of our analysis.
230: Around $T\sim 9.5${\char'27\kern-.3em\hbox{C}}, however, synchronous state is stable both inside
231: and outside the periodic window.  With weak gap junctions the condition
232: for synchronization is so complicated that its intuitive explanation
233: is difficult.
234: 
235: On the other hand, with strong gap junctions, synchronization of MHH
236: neurons is stable over the wide range of temperature~$T$.  In
237: Fig.~\ref{fig:largeg}, we plot the largest transversal Lyapunov
238: exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ as a function of $g$ for various values of
239: temperature~$T$.  Eqs.~(\ref{eq:dave}) and (\ref{eq:dtilde}) show that
240: when parameter~$g$ takes the zero value, transversal Lyapunov
241: exponents~${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ take the same value as tangential Lyapunov
242: exponents~${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}$.  Therefore, synchronous state in
243: Fig.~\ref{fig:largeg} is periodic for $T=6.5,\ 7,\mbox{ and }11.9$~({\char'27\kern-.3em\hbox{C}})
244: and chaotic for $T=7.5,\ 11,\mbox{ and }12.1$~({\char'27\kern-.3em\hbox{C}}).  In these six
245: temperature, the behavior of MHH neurons are quite different from one
246: another.  However, all ${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ take negative values if we
247: increase the strength of gap junctions beyond $g=0.05$.
248: In all the temperature we investigate ($5{\char'27\kern-.3em\hbox{C}}<T<15{\char'27\kern-.3em\hbox{C}}$), we find a
249: certain value of $g$ beyond which ${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ always take a negative
250: value.  Irrespective of whether synchronous state is chaotic or
251: periodic, strong gap junctions induce synchronization of neurons.
252: 
253: In summary, we have studied synchronous state of a large ensemble of
254: modified Hodgkin-Huxley (MHH) neurons assuming gap junctions among
255: neurons.  For the general mean-field dynamics in Eq.~(\ref{eq:general}),
256: we have evaluated $N\times n$-dimensional deviation to define tangential
257: Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\parallel$}}_{l}}\}}$ and transversal Lyapunov exponents~${\{ {\lambda^{\mbox{\tiny $\perp$}}_{l}}\}}$.
258: In Fig.~\ref{fig:result}{(b)}, we have investigated characteristic of
259: synchronous state of MHH neurons by the largest tangential Lyapunov
260: exponent~${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$. In Fig.~\ref{fig:result}{(c)}, we have elucidated
261: stability of this synchronization by the largest transversal Lyapunov
262: exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$.  With weak gap junctions $g=0.02$~{$\mbox{mS/cm}^2$}, stability of
263: synchronization of MHH neurons shows rather complicated change with
264: temperature as shown in Fig.~\ref{fig:result}{(c)}.  However, in
265: Fig.~\ref{fig:largeg} and so on, we have found that with strong gap junctions,
266: synchronous state is stable over the wide range of temperature.
267:   The
268: strong gap junctions induce synchronization both in periodic and chaotic
269: neurons, and that implies a pivotal role of gap
270: junctions in synchronization of the large number of neurons
271: 
272: It should be emphasized that the computational cost for Lyapunov
273: exponents of four-dimensional MHH equations is not much higher than that
274: of the three-dimensional Lorenz equations.  Even when we assume dozens
275: of ion channels in neuron dynamics, we would be able to calculate both
276: Lyapunov exponents within the acceptable computation time.  When
277: synchronous state is periodic and interactions are infinitesimal ($g\ll
278: 1$), one can use the phase reduction method.  In the present system of
279: MHH neurons, however, ${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ in Fig.~\ref{fig:largeg} fluctuates
280: dramatically as $g$ increases, even if synchronous state is
281: periodic. Since the synchronization property with finite $g$ is quite
282: different from that with infinitesimal $g$, calculation of transversal
283: Lyapunov exponents is indispensable in investigating the present system.  In
284: the real nervous system many types of neurons are interconnected with
285: chemical synapses.  Some authors model the dynamics of chemical synapse
286: in the manner as $I_i=(g/N)\sum_j (v_{rev}-v_i)s_j$, where constant
287: $v_{rev}$ denotes reversal potential and $s_j$ obeys the dynamics
288: $ds_j/dt=\alpha F(v_j)(1-s_j)-\beta s_j$ with the sigmoidal function
289: $F(v)=1/(1+\exp[-(v-\theta_{syn})/2])$\cite{perkel,wang,wang2}.  In this
290: case, synchronous state~${{\bf x}}^\ast$ depends on $g$ since
291: ${{\bf G}}({{\bf x}}^\ast,{{\bf x}}^\ast)$ does not vanish.  Moreover, Jacobi matrices of
292: ${{\bf G}}({{\bf x}}_i,{{\bf x}}_j)$ are not constant but depend on ${{\bf x}}_i$ and ${{\bf x}}_j$.
293: Our analysis is applicable also to such complicated neural networks
294: since their dynamics are written in the form of Eq.~(\ref{eq:general}).
295: Although pulse-coupled neural networks based on threshold-crossing spike
296: timing cannot be written in the form of Eq.~(\ref{eq:general}), we can
297: employ the similar analysis by carrying out the decomposition of linear
298: stability discussed in our previous study\cite{myosioka7}.  In that
299: study, we have evaluated two types of Floquet matrices to show that
300: periodic synchronous state of integrated-and-fire (IF) neurons are
301: stable with only inhibitory chemical synapses.  Interestingly, in the
302: real nervous system, interneurons are found to be connected with
303: inhibitory chemical synapses and gap junctions\cite{whittington2}.  It
304: turns out that networks of interneurons take the extremely ideal
305: structure to induce synchronization of an ensemble of neurons.  The
306: present approach of stability analysis is applicable to a wide class of
307: stability problems in neural networks.  Retrieval state in associative
308: memory neural networks of spiking neurons can be investigated by the
309: similar stability analysis\cite{myosioka6}.  More complicated neural
310: networks including pyramidal neurons and interneurons\cite{kopell} would
311: also be analyzed by the present approach.
312: 
313: \begin{thebibliography}{99}
314: \bibitem{buzsaki3} G. Buzs{\'a}ki, Z. Horv{\'a}th, R. Urioste, J. Hetke, and K.~Wise, Science, 256, 1025 (1992).
315: \bibitem{wang} X.J~Wang and J.~Rinzel, J. Neurosci., 16, 6402 (1996).
316: \bibitem{golomb} D.~Golomb and J.~Rinzel, Phys. Rev. E, 48, 4810 (1993).
317: \bibitem{whittington1} M.A.~Whittington, R.D.~Traub, and J.G.R.~Jeffreys, Nature, 373, 612 (1995).
318: \bibitem{wang2} X.J.~Wang and G.~Buz{\'a}ki, J. Neurosci., 16, 6402 (1996).
319: \bibitem{chow}  C.C.~Chow, J.A.~White, J.~Ritt, and N.~Kopell, J. Comput. Neurosci., 5, 407 (1998).
320: \bibitem{tiesinga} P.H.E~Tiesinga and J.V.~Jose, Network 11, 1 (2000).
321: \bibitem{myosioka7} M.~Yoshioka, Phys. Rev. E, in press.
322: \bibitem{ermentrout} G.~Ermentrout and N.~Kopell, J. Math. Biol., 29, 195, (1991).
323: \bibitem{hansel} D.~Hansel, G.~Mato, and C.~Meunier, Neural Comput., 7, 307 (1995).
324: \bibitem{kuramoto} Y.~Kuramoto, Chemical oscillations, waves, and turbulence (Springer-Verlag 1984).
325: \bibitem{fujisaka} H.~Fujisaka and T.~Yamada, Prog. Theor. Phys. 69, 32 (1983).
326: \bibitem{kaneko} K.~Kaneko, Physica D, 77, 456 (1994).
327: \bibitem{maistrenko} Y.~Maistrenko, T.~Kapitaniak, and P.~Szuminski, Phys. Rev. E, 54, 3285 (1996).
328: \bibitem{pikovsky} A.~Pikovsky, O.~Popovych, and Yu.~Maistrenko, Phys. Rev. Lett., 87, 044102 (2001).
329: \bibitem{braun} H.A.~Braun, M.T.~Huber, M.~Dewald, K.~Sch{\" a}fer, and K.~Voigt, Int. J. Bifurcation Chaos
330: 	Appl. Sci. Eng. 8, 881 (1998).
331: \bibitem{feudel} U.~Feudel, A.~Neiman, X.~Pei, W.~Wojtenek, H.~Braun,
332: 	M.~Huber, and F.~Moss, Chaos, 10, 231 (2000). According to this 
333: 	paper we assume the four-dimensional MMH equations of the form
334: 	$\dot{v}=(-I_l-I_d-I_r-I_{sd}-I_{sr})/c,\ 
335: 	I_l=g_l(v-v_l),\
336: 	I_d=\rho g_d\alpha_{d\infty}(v-v_d),\
337: 	\alpha_{d\infty}=(1+\exp[-s_d(v-v_{0d})])^{-1},\
338: 	I_r=\rho g_r\alpha_{r}(v-v_r),\
339: 	\dot{w}_1=\dot{\alpha}_{r}=\phi(\alpha_{r\infty}-\alpha_{r})/\tau_r,\
340: 	\alpha_{r\infty}=(1+\exp[-s_r(v-v_{0r})])^{-1},\
341: 	I_{sd}=\rho g_{sd}\alpha_{sd}(v-v_{sd}),\
342: 	\dot{w}_2=\dot{\alpha}_{sd}=\phi(\alpha_{sd\infty}-\alpha_{sd})/\tau_{sd},\
343: 	\alpha_{sd\infty}=(1+\exp[-s_{sd}(v-v_{0sd})])^{-1},\
344: 	I_{sr}=\rho g_{sr}\alpha_{sr}(v-v_{sr}),\
345: 	\dot{w}_3=\dot{\alpha}_{sr}=\phi(-\eta{}I_{sr}-\theta\alpha_{sr})/\tau_{sr}$
346: 	with temperature-dependent scaling factors
347: 	$\rho=A_1^{(T-T_0)/10}$ and $\phi=A_2^{(T-T_0)/10}$.
348: 	The parameters values are
349: 	$c=1,\
350: 	v_l=-60,\ v_d=v_{sd}=50,\ v_r=v_{sr}=-90,\
351: 	g_l=0.1,\ g_d=1.5,\ g_r=2.0,\ g_{sd}=0.25,\ g_{sr}=0.4,\
352: 	\tau_r=2.0,\ \tau_{sd}=10,\ \tau_{sr}=20,\
353: 	v_{0d}=-25,\ v_{0r}=-25,\ v_{0sd}=-40,\
354: 	s_d=0.25,\ s_r=0.25,\ s_{sd}=0.09,\
355: 	\eta=0.012,\ \theta=0.17,\
356: 	A_1=1.3,\ A_2=3.0,\ T_0=25$.
357: \bibitem{shimada} I.~Shimada and T.~Nagashima, Prog. Theor. Phys. 61, 1605 (1979).
358: \bibitem{milnor} J. Milnor, Commun. Math. Phys. 99, 117 (1985).
359: \bibitem{whittington2} M.~Whittington and R.D.~Traub, Trends Neurosci. 26, 676 (2003).
360: \bibitem{myosioka6} M.~Yoshioka, Phys. Rev. E, 66, 061913 (2002).
361: \bibitem{perkel} D.H.~Perkel, B.~Mulloney, R.W.~Budelli, Neurosci. 6, 823 (1981).\bibitem{kopell} N.~Kopell, G.B.~Ermentrout, M.A.~Whittington, and R.D.~Traub, Proc. Natl. Acad. Sci. USA, 97, 1867 (2000).
362: \end{thebibliography}
363: 
364: \begin{figure}[h]
365:  {}{{}}{}{}
366: \begin{center}
367: \includegraphics[scale=1.0]{001.eps}
368: \end{center}
369:  \caption{{(a)}\ The interspike intervals (ISIs) in stationary state of
370:  an isolated single MHH neuron are plotted as a function of
371:  temperature~$T$.  The ISI distribution of synchronous state of $N$
372:  neurons ${{\bf x}}^\ast$ is the same as a isolated single neuron since the
373:  interaction term in Eq.~(\ref{eq:sync}) vanishes in the present system.
374:  {(b)}\ The largest tangential Lyapunov exponent~${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$, which
375:  characterizes synchronous state~${{\bf x}}^\ast$ described in {(a)}, is
376:  plotted. In the present system, ${\lambda^{\mbox{\tiny $\parallel$}}_{1}}$ is independent of $g$.
377:  {(c)}\ The largest transversal Lyapunov exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ is plotted
378:  for $g=0.02$~{$\mbox{mS/cm}^2$}. When ${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ takes a negative value,
379:  synchronization is stable.  The sampling points are $T=5+0.025\times
380:  k\mbox{({\char'27\kern-.3em\hbox{C}})}\ (k=0,1,2,\ldots)$. }\label{fig:result}
381: \end{figure}
382: 
383: \begin{figure}[h]
384:  {}{{}}{}{}
385: \begin{center}
386: \includegraphics[scale=1.4]{002.eps}
387: \end{center}
388:  \caption{The results of numerical simulations of 100 MHH neurons
389:  with $g=0.02$~{$\mbox{mS/cm}^2$} are plotted for {(a)}\
390:  $T=11.9${\char'27\kern-.3em\hbox{C}}\ and {(b)}\ $T=12.1${\char'27\kern-.3em\hbox{C}}.
391:  Each dot represents spike timing in stationary state realized after $t=1.0\times 10^7$~msec.}\label{fig:simulation}
392: \end{figure}
393: 
394: \begin{figure}[h]
395:  {}{{}}{}{}
396: \begin{center}
397: \includegraphics[scale=1.4]{003.eps}
398: \end{center}
399:  \caption{The largest transversal Lyapunov exponent~${\lambda^{\mbox{\tiny $\perp$}}_{1}}$ is plotted as a
400:  function of $g$ for various values of temperature~$T$. The numbers in the
401:  figure indicate temperature~$T$.}\label{fig:largeg}
402: \end{figure}
403: 
404: \end{document}
405: